3 ms·

This article is almost entirely speculation, and I don't believe it. 1. ChatGPT use declines 2. users complain about ‘dumber’ answers Correlation is not caus

This article is almost entirely speculation, and I don't believe it. 1. ChatGPT use declines 2. users complain about ‘dumber’ answers Correlation is not causation, and a significant fraction of GPT users are students, many/most of whom are currently on summer break.

vinyl7 3y ago
Seems like a pretty good business model to me. Quietly dumb down the AI, then release version N+1 that is smarter and charge $X + 15% for it
- sebzim4500 3y ago
  Notably only works when there is no competition. Thankfully anthropic exists so we don't need to rely on google getting their shit together.
  - regularfry 3y ago
    They'll be marketed on the basis of individual personalities before long. I think it's inevitable, it means each one forms its own natural monopoly.
    - 3cats-in-a-coat 3y ago
      Most people using AI don't give a damn about their personalities, but how well they do work. Imagine if everyone's quirky personality provided them with job safety. In a small company where the boss is sentimental and what not, it may work. In a serious business, having a personality might rather get you fired.
      - regularfry 3y ago
        We already have entire industries based around uniqueness of personality. That's why "casting director" exists as a job. People are also profoundly irrational around anything anthropomorphic, to the extent of being unable to consistently order how well humans "do work" without succumbing to biases. I think it's inevitable not because it makes for a better AI product, but because humans are reliably easy to fool.
        3cats-in-a-coat 3y ago
        I think you're overestimating the percent of GPT users who need it as a synthetic replacement of a friend or a partner. The people who do this are fundamentally looking for relief in the wrong place. The thing doesn't even have memory beyond the immediate chat log. I know online we can easily end up pockets of people to whom every advancement in robotics is about love dolls and every advancement of genetics is about creating catgirls, but believe me these are tiny echo chambers. Humanity at large doesn't process AI in this way at all. They're using it to write homeworks, to do work, and research. Talking about "casting directors" suggests there's something fundamentally off about how you see AI. They're not actors hired to perform a play, or something.
        regularfry 3y ago
        I'm not talking about synthetic replacement. That's a red herring. Humans will make decisions about which LLM to use based on factors entirely unrelated to direct task performance. They may be conscious of this irrationality, but it is more likely that they will not. My prediction is that AIs (in the LLM, homework-writing, prompt-in, response-out sense) will be consciously marketed on the basis of those non-task-specific factors. > Talking about "casting directors" suggests there's something fundamentally off about how you see AI. They're not actors hired to perform a play, or something. No, you've missed my point. Actors are hired because they have a monopoly on themselves. Musicians even more so. LLMs will be the same. Beyond a base level of competence, there will be "competition" between LLMs from different companies in the same sense there is "competition" between operatic baritones. You don't buy a ticket to see Bryn Terfel because he can hit a middle C.
        3cats-in-a-coat 3y ago
        This kind of dynamic you’re describing is inherently based on limited exposure. Tom Cruise wouldn’t have his star power if there were millions of copies all over the world all around you.
        3cats-in-a-coat 3y ago
    - cyanydeez 3y ago
      Not personality. They'll be properly constructed to only understand say, pokemon or star trek. Most people don't care about personality.
- bhouston 3y ago
  That is similar to the old iPhones gets slower after the release of a newer faster iPhone theory, a theory which turned out to be true: https://www.cbc.ca/news/business/apple-will-pay-up-to-500-million-us-to-settle-slow-iphone-lawsuit-1.5482313 https://www.cbc.ca/news/business/apple-will-pay-up-to-500-mi...
  - LeoPanthera 3y ago
    You have enormously misrepresented that story. It wasn't true as anyone actually believed. The processor throttled when the battery could no longer deliver the current necessary to drive it at full speed. It had nothing to do with new iPhones.
YeGoblynQueenne 3y ago
>> This article is almost entirely speculation, and I don't believe it. In support of your skepticism, I'm quoting here the main passage where the article is speculating about the change in performance: A common consensus was that GPT-4 was able to generate outputs faster, but at a lower level of quality. Peter Yang, a product lead for Roblox, took to Twitter to decry the bot’s recent work, claiming that “the quality seems worse”. One forum user said the recent GPT-4 experience felt “like driving a Ferrari for a month then suddenly it turns into a beaten up old pickup”. So all we have to go with is "a common consensus", what it "seems", what someone "felt", and so on. Nothing concrete at all. Which shouldn't be a surprise. Empirical investigations of large language models can either be cheap (in time and money) and perfunctory, or systematic and expensive, and most users don't have the budget, or the inclination, for systematic evaluations. Users form an impression about the performance of LLMs from little experience and then they change their minds, again from a little experience. That's nothing to base anything on.
indymike 3y ago
> significant fraction of GPT users are students, many/most of whom are currently on summer break What does this imply for how ChatGPT is being used by students?
- LeoPanthera 3y ago
  It implies that they aren't using it at all while not at school, which is why usage has declined.

vinyl7 3y ago

Seems like a pretty good business model to me. Quietly dumb down the AI, then release version N+1 that is smarter and charge $X + 15% for it

sebzim4500 3y ago
Notably only works when there is no competition. Thankfully anthropic exists so we don't need to rely on google getting their shit together.
- regularfry 3y ago
  They'll be marketed on the basis of individual personalities before long. I think it's inevitable, it means each one forms its own natural monopoly.
  - 3cats-in-a-coat 3y ago
    Most people using AI don't give a damn about their personalities, but how well they do work. Imagine if everyone's quirky personality provided them with job safety. In a small company where the boss is sentimental and what not, it may work. In a serious business, having a personality might rather get you fired.
    - regularfry 3y ago
      We already have entire industries based around uniqueness of personality. That's why "casting director" exists as a job. People are also profoundly irrational around anything anthropomorphic, to the extent of being unable to consistently order how well humans "do work" without succumbing to biases. I think it's inevitable not because it makes for a better AI product, but because humans are reliably easy to fool.
      - 3cats-in-a-coat 3y ago
        I think you're overestimating the percent of GPT users who need it as a synthetic replacement of a friend or a partner. The people who do this are fundamentally looking for relief in the wrong place. The thing doesn't even have memory beyond the immediate chat log. I know online we can easily end up pockets of people to whom every advancement in robotics is about love dolls and every advancement of genetics is about creating catgirls, but believe me these are tiny echo chambers. Humanity at large doesn't process AI in this way at all. They're using it to write homeworks, to do work, and research. Talking about "casting directors" suggests there's something fundamentally off about how you see AI. They're not actors hired to perform a play, or something.
        regularfry 3y ago
        I'm not talking about synthetic replacement. That's a red herring. Humans will make decisions about which LLM to use based on factors entirely unrelated to direct task performance. They may be conscious of this irrationality, but it is more likely that they will not. My prediction is that AIs (in the LLM, homework-writing, prompt-in, response-out sense) will be consciously marketed on the basis of those non-task-specific factors. > Talking about "casting directors" suggests there's something fundamentally off about how you see AI. They're not actors hired to perform a play, or something. No, you've missed my point. Actors are hired because they have a monopoly on themselves. Musicians even more so. LLMs will be the same. Beyond a base level of competence, there will be "competition" between LLMs from different companies in the same sense there is "competition" between operatic baritones. You don't buy a ticket to see Bryn Terfel because he can hit a middle C.
        3cats-in-a-coat 3y ago
        This kind of dynamic you’re describing is inherently based on limited exposure. Tom Cruise wouldn’t have his star power if there were millions of copies all over the world all around you.
        3cats-in-a-coat 3y ago
  - cyanydeez 3y ago
    Not personality. They'll be properly constructed to only understand say, pokemon or star trek. Most people don't care about personality.
bhouston 3y ago
That is similar to the old iPhones gets slower after the release of a newer faster iPhone theory, a theory which turned out to be true: https://www.cbc.ca/news/business/apple-will-pay-up-to-500-million-us-to-settle-slow-iphone-lawsuit-1.5482313 https://www.cbc.ca/news/business/apple-will-pay-up-to-500-mi...
- LeoPanthera 3y ago
  You have enormously misrepresented that story. It wasn't true as anyone actually believed. The processor throttled when the battery could no longer deliver the current necessary to drive it at full speed. It had nothing to do with new iPhones.

YeGoblynQueenne 3y ago

>> This article is almost entirely speculation, and I don't believe it. In support of your skepticism, I'm quoting here the main passage where the article is speculating about the change in performance: A common consensus was that GPT-4 was able to generate outputs faster, but at a lower level of quality. Peter Yang, a product lead for Roblox, took to Twitter to decry the bot’s recent work, claiming that “the quality seems worse”. One forum user said the recent GPT-4 experience felt “like driving a Ferrari for a month then suddenly it turns into a beaten up old pickup”. So all we have to go with is "a common consensus", what it "seems", what someone "felt", and so on. Nothing concrete at all. Which shouldn't be a surprise. Empirical investigations of large language models can either be cheap (in time and money) and perfunctory, or systematic and expensive, and most users don't have the budget, or the inclination, for systematic evaluations. Users form an impression about the performance of LLMs from little experience and then they change their minds, again from a little experience. That's nothing to base anything on.

indymike 3y ago

> significant fraction of GPT users are students, many/most of whom are currently on summer break What does this imply for how ChatGPT is being used by students?

LeoPanthera 3y ago
It implies that they aren't using it at all while not at school, which is why usage has declined.